Adaptive encoding of zoomable video streams based on user access pattern

摘要

Zoomable video allows users to selectively zoom and pan into regions of interest within the video for viewing at higher resolutions. Such interaction requires dynamic cropping of RoIs on the source video. We have previously explored two different ways of encoding and transmitting video to support dynamic RoI cropping: (i) Monolithic streaming uses a standard video encoder to encode the video. When an RoI is requested, the bits belonging to the RoI along with other bits required to decode the RoIs (due to encoding dependencies) are transmitted. (ii) Tile streaming divides regions in the standard video into rectangular tiles that are encoded independently. The tiles that intersect with a requested RoI are transmitted. In this paper, we consider how the bandwidth needed to transmit the RoIs can be reduced by carefully encoding the source video for each of the two encoding schemes. The goal is to support bandwidth efficient compressed domain RoI cropping in the context of virtual zoom and pan by tuning encoder parameters. Our key idea is to exploit user access patterns to the RoIs, and encode different regions of the video with different encoding parameters based on the popularity of the region. We show that our encoding method can reduce the expected bandwidth by up to 43% in the test video sequence which we have used.